RAG Knowledge Base Systems
A curated list of RAG (Retrieval-Augmented Generation) frameworks, platforms, and vector databases. Last updated: 2025.
RAG Frameworks
Developer-facing libraries for building RAG pipelines.
| Project | Company | Stars | License | Description |
|---|---|---|---|---|
| LangChain | LangChain Inc. (US, a16z) | 90k+ | MIT | De facto standard LLM app framework. Modular chain architecture for RAG; LangGraph for agentic RAG, LangSmith for observability. |
| LlamaIndex | LlamaIndex Inc. (US) | 38k+ | MIT | Data framework purpose-built for RAG. 100+ connectors, hierarchical retrieval, multimodal RAG. Best for document Q&A. |
| Haystack | deepset (Germany) | 17k+ | Apache 2.0 | Enterprise end-to-end LLM/RAG pipeline framework. v2 redesigned in 2024; built-in evaluation, production-ready. |
All-in-One Knowledge Base Platforms
| Project | Company | Stars | License | Description |
|---|---|---|---|---|
| Dify | LangGenius (China) | 45k+ | Apache 2.0 | Visual drag-and-drop LLM app platform with built-in RAG pipeline and knowledge base management. Supports 100+ models. |
| RAGFlow | InfiniFlow (China) | 30k+ | Apache 2.0 | RAG engine based on deep document understanding. Excels at complex PDFs, tables, and charts. Supports citation tracing. |
| FastGPT | labring (China) | 20k+ | Apache 2.0 | LLM knowledge base platform with visual Flow workflow orchestration. TypeScript/Next.js, team collaboration support. |
| AnythingLLM | Mintplex Labs (US) | 30k+ | MIT | Full-stack desktop/server LLM app. Turns any document into a RAG chatbot. Multi-user, local+cloud LLM, multi-vector-DB. |
| MaxKB | FIT2CLOUD (China) | — | Apache 2.0 | Enterprise RAG Q&A bot supporting multiple LLM backends. From the 1Panel/JumpServer ecosystem. |
| Quivr | Quivr HQ (France) | 36k+ | Apache 2.0 | "AI second brain" personal knowledge assistant. Multi-format RAG retrieval, cloud and local deployment. |
| PrivateGPT | Zylon (privacy AI) | 54k+ | Apache 2.0 | 100% local, privacy-first document Q&A. Fully offline. One of the earliest viral local RAG projects. |
| Kotaemon | Cinnamon AI (Vietnam/Japan) | 18k+ | Apache 2.0 | Clean, customizable document RAG chat tool with Gradio UI. Multi-vector retrieval, citation display, GraphRAG integration. |
| Onyx (Danswer) | Onyx (YC W23) | 13k+ | MIT | Connects 50+ sources (Confluence/Notion/Slack). Self-hosted RAG enterprise Q&A platform, any LLM. |
| WeKnora | Tencent (China) | — | Apache 2.0 | Enterprise knowledge platform open-sourced by Tencent. "Wiki Mode" productizes Karpathy's LLM-as-Compiler pattern. Hybrid retrieval (vector + BM25 + knowledge graph), multi-tenant RBAC, built-in Agent, IM integration. |
Vector Databases
| Project | Company | Stars | License | Description |
|---|---|---|---|---|
| Chroma | Chroma (US, YC) | 16k+ | Apache 2.0 | Lightweight embedded vector database. Easiest to get started with for RAG prototyping. |
| Weaviate | Weaviate B.V. (Netherlands, $50M+) | 11k+ | BSD | Vector database with GraphQL API. Supports hybrid search and multimodal retrieval. |
| Qdrant | Qdrant (Germany, $28M Series A) | 21k+ | Apache 2.0 | High-performance vector search engine written in Rust. Excellent filtering for production RAG. |
| Milvus | Zilliz (US/China, $113M+) | 30k+ | Apache 2.0 | Cloud-native vector database for billion-scale vectors. Commercial: Zilliz Cloud. |
| pgvector | Community (Supabase contributor) | — | PostgreSQL | PostgreSQL vector extension with HNSW/IVFFlat index. Zero-cost vector retrieval if already on Postgres. |
Managed Vector Database Services
| Product | Company | Description |
|---|---|---|
| Pinecone | Pinecone Systems (US, $100M+) | Fully managed serverless vector database. Zero ops, most mature managed RAG storage option. Closed source. |
| Zilliz Cloud | Zilliz (Milvus parent) | Managed cloud service for Milvus. Enterprise SLA and security. |
Graph RAG
| Project | Company | Stars | Description |
|---|---|---|---|
| GraphRAG | Microsoft Research | 19k+ | Graph-based RAG via community summaries. Answers complex multi-hop questions. |
| LightRAG | HKU Data Intelligence Lab | 15k+ | Dual graph-vector index. Supports local/global dual-mode retrieval. Emerged late 2024. |
| UnWeaver | Community | — | Entity-level decomposition + aggregation on top of VectorRAG. Beats GraphRAG in factual correctness without building an explicit knowledge graph. |
| TrustGraph | Community | — | Knowledge graph + RAG platform. Combines structured graph reasoning with vector retrieval. |
Vector-Free RAG
Alternative retrieval approaches that don't rely on embedding similarity.
| Project | Company | Stars | Description |
|---|---|---|---|
| PageIndex | VectifyAI | 29k+ | Semantic summary tree replaces vector database. Retrieval = reasoning via LLM tree search. 98.7% accuracy on FinanceBench. |
| Memvid | Community | — | Single .mv2 file replaces the entire RAG pipeline and vector database. Sub-millisecond retrieval. |
| OhMyGo | Community | — | Go-based AI knowledge platform integrating chat, RAG Q&A, image recognition, and speech synthesis. |
| he-wiki-rag | Community | — | Chapter-tree indexing preserves Markdown H1/H2/H3 hierarchy to avoid semantic loss from fixed-length chunking. Vector + BM25 + Cross-Encoder Rerank pipeline. |
Traditional Search Engines (RAG-extended)
| Product | Company | Description |
|---|---|---|
| Elasticsearch | Elastic N.V. (NYSE: ESTC) | Enterprise search extended with ESRE and ELSER sparse vectors. Mature hybrid search (BM25 + vector) at billion-scale. |
| OpenSearch | AWS (Amazon) | AWS open-source Elasticsearch fork. Deep Amazon Bedrock integration. Neural Sparse Retrieval for cloud-native RAG. |